Competitive Compensation Contracts

Tore Ellingsen and Eirik Gaard Kristiansen

Competitive Compensation Contracts∗

Tore Ellingsen† Eirik Gaard Kristiansen‡

September 25, 2016

Abstract

We develop a model of equilibrium compensation contracts in a competitive

market for heterogeneous workers. The eventual value of a match depends on the

fortunes of the employer’s industry, and turnover could be efficiency-enhancing. Ap-

plying the theory of contracts as reference points, we demonstrate that ex ante con-

tracting entails better insurance than does ex post renegotiation and that predicted

contract terms closely resemble observed terms. For example, salaries constitute

lower bounds on pay, salaries are combined with non-indexed stock options when

industry conditions are uncertain and employees possess much portable industry-

specific human capital, severance pay is discretionary, and hedging of stock-options

is permitted.

JEL Codes: J33, D86

Keywords: Compensation contracts, turnover, performance pay

∗Thanks to Carsten Bienz, Mike Burkart, James Dow, Mariassunta Giannetti, Robert Gibbons, BengtHolmstrom, Hans Hvide, Daniel Metzger, Espen Moen, Trond Olsen, Marco Ottaviani, Patrick Rey, PerStromberg, Steve Tadelis, Marko Tervio, and Joel Watson for helpful discussions and several anony-mous referees for comments on the paper’s precursor. Financial support from the Torsten and RagnarSoderberg Foundation and The Finance Market Fund is gratefully acknowledged.†Department of Economics, Stockholm School of Economics, Box 6501, S–11383 Stockholm, Sweden.

E-mail: gte@hhs.se‡Norwegian School of Economics, Helleveien 30, N–5045 Bergen, Norway.

E-mail: eirik.kristiansen@nhh.no

1

1 Introduction

There is a widespread suspicion that top managers and other key personnel are overpaid.

One reason for the suspicion is that such employees are often lavishly rewarded when the

firm is lucky, yet not correspondingly penalized when the firm is unlucky (e.g., Hermalin

and Weisbach, 1998, Bertrand and Mullainathan, 2001, Garvey and Milbourn, 2006, and

Bell and van Reenen, 2016). Indeed, high salaries regularly constitute lower bounds on

pay; even when these employees quit or are fired, they frequently receive discretionary

severance pay that compensate for the anticipated pay reduction in subsequent employ-

ment (e.g., Yermack, 2006).1 Here, we argue that asymmetric rewards and discretionary

severance pay, as well as several other controversial features of common compensation

contracts, are not necessarily evidence of corruption or inept compensation committees.

Instead, these compensation practices may be natural outcomes of fierce competition for

scarce talent, as previously suggested by, among others, Rosen (1992), Holmstrom and

Kaplan (2003), Oyer (2004), and Hubbard (2005).

To articulate our argument, we construct a model of a competitive labor market com-

prising both specialists and non-specialists. A specialist is defined as a worker whose skills

are specific to an industry and possibly even to a firm. Specialists may be craftsmen,

traders, or managers – as long as their particular craft or managerial skill is sufficiently

specific. In the model, industry-wide shocks do not affect the market value of non-

specialists, who thus receive a fixed salary in equilibrium. However, the market value

of specialists will be volatile. Therefore employers will design compensation contracts to

shield their employed specialists against movements in the market value of their services,

while facilitating efficient recruitment and turnover. When providing insurance for spe-

cialists, two factors are crucial. On the one hand, specialists’ firm-specific skills constrain

mobility. On the other hand, an employed specialist may be unable to commit to stay

with the employer or to compensate the employer for leaving in the presence of attractive

outside options. Firm-specific skills facilitate risk-sharing, whereas limited commitment

in combination with economic fluctuations hinder it.

When specialists are sufficiently immobile and the industry is sufficiently stable, the

equilibrium contract is a flat salary. This contract provides maximal income insurance.

But if mobility is greater or industry conditions are less stable, retention of a specialist

requires higher pay in good states. A crucial question is whether such flexibility could be

efficiently provided through renegotiation. We argue that the answer is negative. Even if

parties have complete information and are always able to find ex post acceptable arrange-

1For an extensive critique of excessive pay based on these and related observations, see Bebchuk,Fried, and Walker (2002) and Bebchuk and Fried (2009).

2

ments, these arrangements are unlikely to be ex ante optimal. Specifically, we employ a

version of the “contracts as reference points” theory pioneered by Hart and Moore (2008)

and Hart (2009). The theory has two central components. First, a party that considers

the contract to be unfair will react by under-performing (engage in shading). Second,

the fairness of a contract is ascertained according to how it splits the surplus as viewed

from the date of agreement. Thus, the terms of a mutually agreed contract might be

considered fair ex post by a party that was well treated ex ante, even if the terms turn

out to be relatively unfavorable in the realized state of the world. On the other hand, the

same terms might be considered unfair if they are reached through renegotiation after

the state has materialized. For experimental evidence that contracts serve as reference

points in this way, see, e.g., Fehr, Hart, and Zehnder (2009, 2011, 2015) and Bartling and

Schmidt (2015).2

Since adverse responses to an unfair renegotiation outcome are only relevant when the

relationship continues, the model predicts that there will be rent-sharing when renegotia-

tion produces retention, but not when it produces separation.3 Perhaps most importantly,

even if optimal renegotiation outcomes are such that they never trigger any shading, a

state-contingent contract is always preferable to a sequence of spot-contracts, because

the state-contingent contract enables greater insurance.4

Overall, the model rationalizes the following commonly observed features of special-

ists’ compensation contracts and turnover:5

1. Contracts have a fixed salary component, and total pay is never below this salary.

2. If the specialist’s market value is reliably linked to a verifiable measure, for example

to the firm’s stock market value, the contract offers linear variable pay above a

performance threshold – for example a stock option.

3. The performance component is rarely indexed and thus does not filter out the

impact on pay of macroeconomic or industry-wide shocks.

4. The performance component’s magnitude depends on the specialist’s expected mo-

bility.

2This is not to say that more complete contracts are always beneficial. For experimental evidence ofbenefits from (increased) contractual incompleteness when contracts cannot be perfectly complete andneed to rely on bounded punishment see, e.g., Fehr and List (2004) and Brandts, Charness, and Ellman(2016).

3This prediction is quite different from a model in which bargaining power is a personality trait.4Fudenberg, Holmstrom, and Milgrom (1990) provide general conditions under which an optimal

long-term contract can be mimicked by a sequence of short-term contracts. It is an interesting openquestion whether an analogous result holds under the “contracts as reference points” assumption.

5We provide extensive references to the relevant empirical literature below.

3

5. Employers offer discretionary severance pay.

6. Turnover of specialists is typically related to relatively poor performance of the

whole industry rather than firm-specific performance.

To clarify the mechanisms behind our results, let us give a more detailed intuitive

account of the main steps of the argument. The analysis is conducted within a simple

competitive labor market model. There are several industries. Firms are homogeneous

within an industry. Specialists differ in their industry-specific abilities, but are more sim-

ilar ex ante than after acquiring industry-specific experience. Employees are recruited for

two periods. A contract is signed at the beginning of the first period of employment and

then potentially renegotiated at the beginning of the second period, upon the realization

of industry-wide economic shocks. We assume that all information is symmetrically held.

Employees are more risk-averse than employers. Absent shirking problems or outside

competition, the best compensation contract would thus be a fixed salary and no other

payment. More generally, the ability of firms to commit to a long-term fixed salary –

rather than being confined to offer short-term contracts – is a source of welfare gains in

our model. However, fixed salary contracts imply that workers might be substantially

overpaid in some states of the world, and firms would be tempted to renege on their

promise. Institutions that prevent firms from reneging on their salary promises in bad

states are thus crucial to the implementation of such socially valuable contracts.6

Another obstacle to income insurance through flat salary contracts is that laws against

involuntary servitude prevent the use of penalties. Thus, the employer must reward

employees for staying rather than penalizing them for leaving. At first sight, it would

seem that the employer might then be best off by simply adjusting the salary upwards as

attractive external offers arrive. This is the outcome of the classical model of Harris and

Holmstrom (1982). In their environment of exogenous outside offers, the adjustable salary

mechanism exposes the employee to the minimum income risk subject to free ex post

mobility. However, our environment differs from that of Harris and Holmstrom (1982) in

a crucial way. Due to the accumulation of firm-specific skills in the first period, the worker

will be strictly better matched with the current employer than with any other employer in

the same industry. While the employer will outbid same-industry competitors, contract

renegotiations allow the worker to bargain for some of the firm-specific surplus. If the

employee has any bargaining power at all, the renegotiated compensation will be strictly

above the outside option. Hence, renegotiation imposes greater pay variability than an

optimal contract with explicitly variable pay.

6It is noteworthy that the vast majority of CEO employment contracts are for a fixed term, thuscircumventing the “at-will” legal default; see Schwab and Thomas (2006).

4

Optimal turnover requires that the worker is allowed to leave when the outside match

is better than the original match, and – perhaps less obviously – that the worker is paid

to leave when the current match is bad and worse than the outside match. While a

complete contract would specify the amount of severance pay for each such state of the

world, we argue that it is superfluous to contract on these states in advance. Since the

employee neither has a credible threat to leave and reap a higher income than under

the contract nor can express dissatisfaction by exerting less effort, the employer naturally

has all the bargaining power in severance pay negotiations. Thus, discretionary severance

pay emerges as an optimal device for attaining efficient separation in the case of poor

industry performance. (Discretionary severance pay would not be optimal if we were to

assume that workers have bargaining power under separation as well as under retention.)

In addition to explaining the basic features of employment contracts that we observe

in practice, the model admits a variety of comparative static results, and these results

too are broadly in line with available evidence. For example, the model predicts that the

inter-firm portability of an employee’s skills is the main determinant of the magnitude

of variable pay. The model is also consistent with the observation that the relative

importance of variable pay is frequently greatest in industries with much uncertainty,

another finding that is difficult to square with effort-inducement models.

Although we primarily focus on the case in which outright bonding is illegal, in the

sense that the employer cannot penalize an employee for leaving, we also observe that

there are important segments of the labor market in which bonding can be legally attained

with the help of outside markets – and we argue that some observed pay practices can

be understood in this light. Specifically, the stock-option packages that are often offered

to executives and other key personnel can have desirable bonding properties. Under the

following special circumstances, we show that a stock-option package can even implement

the perfect bonding outcome: (i) the employee will optimally remain matched with the

original employer; (ii) the best outside offer will come from another firm whose match-

value is closely correlated with that of the original employer; and (iii) the employee is

allowed to trade the options in a frictionless financial market.7 The model thus offers

an explanation for why firms permit managers to “undo the incentive effects” of their

performance pay packages through options trading, a phenomenon that is often considered

to constitute definite evidence of managerial rent extraction (Bebchuk, Fried, and Walker,

2002).

7This environment is special, but familiar. Conditions (i) and (ii) are fulfilled by the model of Harrisand Holmstrom (1982).

5

2 Related Literature

The literature on optimal compensation contracts is huge. Yet, besides Harris and Holm-

strom (1982) there are only a handful of closely related contributions. Most of the con-

tracting literature focuses on effort incentives within a relationship, neglecting the issue

of worker mobility. That is, it primarily studies the bilateral relationship between a single

employer and a single employee, with market forces determining the average level of pay

but otherwise playing a subordinate role.8 It is also typically concerned with explaining

the strength of incentives rather than the contracting process and the precise contractual

shape. For example, very few models attempt to explain why contracts comprise both

pay floors and linear performance pay components.

A notable precursor to our work is Holmstrom and Ricart i Costa (1986). There too,

optimal compensation takes the form of an option contract, with the fixed salary being

due to the employee’s risk aversion and the variable pay being due to the employee’s

inability to commit to staying with the current employer when outside opportunities be-

come attractive. However, where Holmstrom and Ricart i Costa emphasize uncertainty

about employee characteristics, we emphasize uncertainty about future market condi-

tions. Therefore, we are able to address many empirical regularities regarding which

their model is silent. For example, we can explain why plain stock options are used to

reward employees whose talents are well known and whose effort does not greatly affect

the value of the firm; in their model, the option is instead tied to what is revealed about

the specific skills of individual employees, for which the stock price is typically a less

precise indicator.9 Another difference is that Holmstrom and Ricart i Costa assume that

mobility barriers are absent. Without any benefit from retention, the magnitude of their

fixed wage component is bounded by the principal’s ability to extract surplus from the

employee through low pay in an initial period. In our model, the magnitude of the fixed

wage is instead largely driven by the size of the mobility barrier, and can thus attain a

more realistic size. Apart from Holmstrom and Ricart i Costa (1986), we are not aware

of any previous model that explains why employees are paid a combination of fixed salary

and non-indexed stock options.10

8See, for example, the surveys of Murphy (1999) and Lazear and Oyer (2012).9In Holmstrom and Ricart i Costa (1986) the option value will be linear in the stock price when there

is a strong impact of worker’s ability on the firm’s value, which is only true for exceptionally importantemployees.

10Models that attempt to explain how option packages vary with firm and market conditions, such asJohnson and Tian (2000), exogenously impose a combination of salary and options. Among previoustheoretical models of compensation contracts that consider the retention motive, Hashimoto (1979) andBlakemore et al. (1987) assume that contracts are piece-wise linear. Oyer (2004) and Giannetti (2011)assume linear contracts. Dutta (2003) derives a linear contract from first principles. All three thus failto account for the lower bound to payments. Finally, Pakes and Nitzan (1983) examine how contracts

6

In other respects, our work is most closely related to Oyer (2004), who also explains

why performance pay is non-indexed and often extends to many employees rather than

just the top management. Like us, Oyer argues that contracts may link pay to the firm’s

performance because both the firm’s performance and the employees’ outside options are

likely to be correlated with industry performance (or more generally with macroeconomic

conditions). If the main purpose of the contract is to ensure retention, the contract should

not be indexed. Furthermore, Oyer argues that firms prefer to contractually commit

to pay for performance rather than negotiating pay increases as outside offers arrive.

However, the details of the latter argument are quite different from ours. Oyer emphasizes

how a long-term contract avoids renegotiation costs ex post, whereas we instead focus

on reduced pay variability. But there are two more important differences that make our

model more general than Oyer’s.11 First, we do not restrict the shape of compensation.

Oyer assumes that contracts are linear. Thus, by construction, he does not account for

the lower bound to compensation that the combination of salary and options implies.

Second, we model the whole labor market rather than just the problem facing a single

firm that tries to retain workers. Among other things, we are thereby able to explain

turnover and severance pay. Our two generalizations thus contribute substantive new

results while at the same time confirming the robustness of Oyer’s main insights.

The role of portable human capital has also recently been studied in the literature

on relational contracting. For example, Kvaløy and Olsen (2012) argues that increased

portability tends to favor individual incentives over group incentives, and will sometimes

yield individual pay that is driven by outside options rather than inside contributions.

Finally, our work is complementary to the literature on matching and compensation

of executives, such as Gabaix and Landier (2008), Tervio (2008), and Edmans, Gabaix,

and Landier (2009). That literature also studies the compensation of employees with

heterogeneous skills. However, the emphasis is on the equilibrium level of pay and on the

matching of specialists within an industry composed of heterogeneous firms, whereas we

instead focus on the shape of the compensation contract and neglect intra-industry firm

heterogeneity.

can be designed to retain research personnel. Their focus is similar to ours, but the contract derivedis generally not linear in performance and it depends on the potential rivalry between old and newemployers.

11This is not to claim that our model nests Oyer’s as a special case. There are additional differences.However, we believe that these are not crucial for the ensuing comparison.

7

3 The Model

There is a set F of risk-neutral firms and a set J of risk-averse workers. Workers are

employable for two periods, but only produce in the second period; the first period can

be thought of as training. Training is costless for both the employer and the employee.

There are n industries. Let I denote the set of industries. We assume that each

firm can only operate projects in a single industry. We assume that a worker can switch

employer (and even industry) between period 1 and period 2.

Tasks, technologies, and industries. There are two kinds of tasks. One kind is sensitive

to employees’ skills, the other kind is not. We refer to these tasks as specialized and

general, respectively. As will become clear, performance-related compensation contracts

will only be valuable in the case of workers that conduct specialized tasks. (Note that this

distinction between specialized and non-specialized tasks is conceptually very different

from the distinction between high-impact and low-impact employees.)

Firms have Leontief production functions.12 Let Jf be the set of workers employed

by firm f , of whom Jsf are assigned to the specialized task and Jgf to the general task.

Firm f in industry i thus produces output

Qfi = min

∑j∈Jg

f

ej,∑j∈Js

f

ejxji

, (1)

where, xji is the productivity of employee j in the firm’s specialized task (all workers have

the same productivity, normalized to 1, in the general task), and ej ∈ R+ is employee j’s

effort.

Uncertainty. Each firm takes the output market price pi as given. Due to variation

in demand, output prices are ex ante uncertain. Let realized prices be denoted p =

(p1, p2, . . . , pn). We assume that p is observable and verifiable. Let h(p) denote the

distribution of p, with H denoting the cumulative distribution.

Worker abilities. Workers have different innate abilities regarding the specialized

tasks. Worker j’s productivity in the specialized task of industry i, xji , is determined by

innate ability and by practice. The innate ability of worker j in industry i is denoted

aji ∈ R. Note that we allow abilities to be negative; specialized tasks are difficult, and

an unqualified employee could do harm. Worker j’s innate ability profile is denoted

aj = (aj1, . . . , ajn). Think of xji = aji as the productivity of worker j with only general

job training in an industry k 6= i. When the worker remains in the original industry,

12We could allow more flexible functional forms, but this would make the analysis more complexwithout substantial new insights.

8

say in industry ι, and with the original employer, the productivity instead increases to

xjι = ajι (1 + κ), with κ > 0, as the worker has obtained specialized training.

While innate abilities are perfectly transferable across firms, we assume that acquired

skills are only partially portable.13 Let θκ denote the acquired skill that is portable to

other firms in industry ι, with θ < 1. The worker’s firm-specific productivity is thus

ajι (1− θ)κ. We assume that there is no portability across industries.

Refer to aj = maxi{aji}

as the worker’s strongest ability.

Definition 1 A worker j is said to be talented if and only if aj > 0.

In other words, only those workers are labelled talented whose employment in a specialized

task might contribute to raise output. Let J denote the set of talented workers. We

assume that this set is relatively small.

Assumption 1∑

j∈J(1 + κ)aj < |J | − |J |.

As will soon be evident, this assumption ensures that there is always a deficit of talented

workers relative to specialized jobs. For simplicity, we assume that non-specialists do not

need to be employed in period 1 in order to be productive in period 2; they simply need

the first period to mature.14 Productivity and its components, talent and training, are

commonly observable.

Worker preferences. Let rj denote the total remuneration that worker j receives.

Workers care about remuneration and effort according to the utility function

U j = u(rj)− c(ej, β). (2)

The function u(·) is increasing and strictly concave.15 As in Holmstrom and Milgrom

(1991), we assume that the effort cost function c(·) is strictly convex in e, initially de-

creasing and eventually increasing. Thus, a worker prefers to exert some positive amount

of effort when employed, even without any material incentive.

The “non-standard” element of the utility function is β, which indicates how well the

worker likes the employer. Here, we assume that β is determined by how generously the

13Portability of human capital is affected both by technology and institutions; for empirical illustrationsand relevant references see, for example, Groysberg, Lee, and Nanda (2009) and Marx, Strumsky, andFleming (2009).

14This assumption saves us from keeping track of the period 1 employment of non-specialists. In amore dynamic model, the natural alternative assumption would be that non-specialists are productivelyemployed in both periods.

15As usual, we imagine that all consumption takes place “at the end.” While it is feasible to modelthe interaction between compensation contracts and intertemporal consumption decisions, tractabilitytends to require strong assumptions. For example, Harris and Holmstrom (1982) assume that utility istime-separable and that workers cannot save or borrow.

9

employer treats the worker in negotiations. Specifically, let β denote the share of available

(quasi-) rents that the the employer offers to the employee.16 But rather than pursuing

a general analysis of reciprocity, we follow the example of Hart and Moore (2008) and

study negative reciprocity only. Specifically, we impose the simplification

U j = u(rj)− c(g(β)ej), (3)

where,

g =

gN if β ≥ β;

gA if β < β;(4)

and 0 < gN < gA. The value gA indicates that the worker is aggrieved. In the tradition

of Hart and Moore (2008) and Fehr, Hart and Zehnder (2009, 2011), we assume that the

worker feels aggrieved if and only if the employer, when negotiating compensation, offers

the worker a fraction of the relevant gains from trade that is smaller than some level

β > 0.17

We choose the function c such that the minimizer of c(gNe) is eN = 1. It follows that

the minimizer of c(gAe), call it eA, satisfies eA < 1.18 The following assumption guarantees

that the employer will want to avoid triggering grievance.

Assumption 2 β < 1− eA.

Profits. From (1) it follows that a profit-maximizing firm will minimize costs by

setting ∑j∈Jg

f

ej =∑j∈Js

f

ejxji .

Since there is never any quasi-rent to be shared with individual employees in the general

task, there is no reason for grievance; hence we anticipate that ej = 1 for employees in

the general task. Let wg denote the compensation of non-specialists and wj denote the

compensation of specialist j. The profit of a firm is then

πfi = (pi − wg)∑j∈Js

f

ejxji −∑j∈Js

f

wj. (5)

Match surplus. From now on, we drop scripts j and f whenever it is clear that we

talk about a single specialist. Let it denote the industry of a worker’s employer in period

16A natural general assumption would be c′′e,β < 0.17In the original Hart-Moore model, parties feel aggrieved if they do not obtain the maximal outcome

that is feasible under the contract. Here, that would mean that workers are aggrieved if they do notreceive all the gains from trade; i.e., β = 1.

18Consider for example c = (1− ge)2, with gN = 1 and gA = 2.

10

t. Throughout, when we refer generically to i1, we let i1 = ι.

Using (5), we define the (gross) surplus that is generated by a matched specialist, and

how it depends on where the specialist spends period 2. If the specialist remains in the

initial industry (industry ι), the surplus is

Sι(pι) = (pι − wg)eaι(1 + κ) (6)

if the specialist remains with the original employer and

Sι(pι) = (pι − wg)eaι(1 + θκ) (7)

if the specialist moves to another employer in the same industry. For industries i 6= ι,

the specialist generates a match surplus

Si(pi) = (pi − wg)eai. (8)

Let

Sι(p) = max{maxi 6=ι{Si}, Sι, 0}, (9)

denote the surplus associated with the best possible matching of specialist j in state p,

and let

Sι(p) = max{maxi 6=ι{Si}, Sι, 0}, (10)

denote the best outside option for specialist j in state p.19 Finally, let

Tι(p) = Sι − Sι (11)

denote the gain from trade that is reaped if the specialist remains with the original

employer instead of leaving.

Define the surplus-maximizing task assignment in period 2 given assignment ι in

period 1:

i∗2(ι) = arg maxi

(pi − wg)(1 + 1ιk)ai, (12)

where 1ι is an indicator function taking value 1 if the specialist is assigned to the same

firm in both periods and 0 otherwise. The expected surplus, given that the worker is

initially assigned to industry ι and subsequently assigned to i∗2(ι), is E[Sι]. (Only the

prices p are stochastic, so we suppress the argument on E.) The optimal assignment in

19Here we assume that unemployment is the outcome in case a talented worker ends up with nothaving valued specialized skills; alternatively, we might have assumed that talented workers are preferredin general jobs, in which case the relevant pay would be wg instead.

11

period 1 maximizes the expected surplus in the second period,

i∗1 = arg maxιE[Sι]. (13)

Let the expected surplus given optimal assignments in both periods be denoted

S∗ = E[Si∗1]. (14)

Contracts. Throughout, we assume that effort is not explicitly contractible. However,

assignment and pay is contractible to variable degrees. Let jt ∈ F denote the employer of

worker j in period t. By definition, a contract accepted in period t implies employment

in period t. For the period 1 contract, we consider the following five cases, in decreasing

order of contracts’ complexity and the parties’ strength of commitment.

1. Unconstrained contracts. An unconstrained contract specifies a wage w(a, p, j2) ∈R. Note that pay can be negative, and that it may be positive even in case the

worker does not remain with the initial employer. We think of unconstrained con-

tracts as a benchmark rather than a realistic description of actual contracts.

2. Bounded contracts. A bounded contract specifies a wage wb(a, p, j2) ≥ λ. Thus, the

difference between a bounded contract and an unconstrained contract is that the

bounded contract has a lower bound λ on pay. When λ = 0 (as we will assume), this

can be interpreted as a no-servitude law; in our context the lower bound matters

because it precludes the employer from imposing fines (transfer fees) on workers

who leave the firm.

3. Simple performance contracts. The simple performance contract is a bounded con-

tract that specifies a state-contingent wage wc(a, pf , j2 = j1) that is non-negative if

the worker remains with the firm and zero otherwise. In addition to being bounded

below, the wage depends only on the profitability of the firm and on whether the

worker is assigned to the firm or not.

4. Fixed-salary contracts. In this case, the firm commits to make the same payment,

wa, in all states – as long as the worker remains with the firm in period 2; otherwise

the pay is zero.

5. Spot contracts. In this case, the firm offers only a short-term contract at date 1.

That is, it commits to pay ws in all states, regardless of whether the worker stays

in period 2 or not.

12

In addition to the contracts that are agreed before the resolution of uncertainty, we

allow renegotiation and new contracts to be signed after p is realized. Since the state is

known at this stage, the renegotiation offer only depends on where the employee works.

In general, we allow the renegotiation offer to be any real-valued function wr(i2). For

the cases 3 and 4, this implies that renegotiation enables outcomes that could not be

specified in the initial contract. For example, the renegotiation offer might pay out

positive amounts even when the worker does not stay.

Timing and information. Our results are not very sensitive to the exact timing of

offers, but the complexity of the analysis is. We choose the timing that admits the most

straightforward analysis.

1. Contracting:

(a) Firms propose contracts to as many talented workers they like.

(b) Each talented worker receiving an offer decides one and only one offer to accept.

2. Competition and turnover:

(a) The state p is realized.

(b) Firms make new offers to their current specialists.

(c) Firms make new offers to other talented workers.

(d) Each talented worker decides which of the new offers to accept (thus becoming

a specialized employee).

(e) All unmatched workers supply non-specialized services; the wage for the gen-

eral task, wg, is determined in a competitive (Bertrand) fashion.

3. Workers exert effort and are paid:

(a) Each specialist decides how much effort to exert.

(b) All workers are paid according to the agreed contracts.

We assume that the history is common knowledge among all players. In particular, at

Stage 2c firms know which offers were made at Stage 2b.

4 Analysis

We seek to characterize the subgame-perfect equilibrium outcome(s) of this game under

various assumptions about the set of available contracts. However, before doing so, let

us characterize the optimal outcomes.

13

4.1 Optimal outcomes

It is easily seen that a first-best outcome requires an assignment of each talented worker

that maximizes that worker’s expected surplus E[S] while simultaneously providing com-

plete insurance.

Proposition 1 (First-best) An outcome is first-best only if: (i) the assignment of each

talented worker satisfies (i∗1, i∗2) and (ii) each talented worker’s remuneration is fixed, i.e.,

r is independent of p.

Proof. See Appendix.

This result borders on the trivial. Since firms are risk-neutral, workers ought to be

maximizing expected output, with firms carrying all the risk.

The harder question concerns the social arrangements that are needed to sustain such

a high degree of insurance. In particular, under what conditions will a competitive market

provide it? We now seek to answer this question.

4.2 Worker effort and contract renegotiation

In order to characterize subgame-perfect equilibria, the analysis starts at the last stage

and moves backwards.

Stage 3. If the prevailing contract was determined by an offer that granted the em-

ployee at least a fraction β of the available surplus at the time the contract was signed, the

employee exerts effort e = 1; otherwise, e = e. (As we shall see, if the prevailing contract

was signed at Stage 1, the employee will usually have been offered all the available surplus

at that stage. Likewise, if an employee switched employer at Stage 2, the specialist will

have obtained all the surplus. Thus, shading of effort will only be a potential concern if

the current contract was proposed during renegotiation with the initial employer at Stage

2.)

Stage 2e. Due to Assumption 1, we know that there is an excess of non-specialist

workers regardless of the contracting outcome involving talented workers. Hence, the

unique competitive wage for the general task equals the workers’ reservation value. That

is, wg = 0.

Stage 2d. Since workers only care about consumption, each talented worker accepts

the offer (one of the offers) that yields the highest total pay.

Stage 2c. At this stage, there is Bertrand competition for talented workers. Consider

any talented worker j who either does not hold an offer or who holds an offer that is

below Sι. Then, by the standard Bertrand logic, in any subgame-perfect pure-strategy

equilibrium, at least two firms will make offers of exactly Sι.

14

Stage 2b. Consider an incumbent specialist in firm f . Let winc = w(p, f) − w(p, f ′)

denote the net pay that is specified by the incumbent employer’s contract if the employee

stays rather than leaves. (If there is no long-term contract, we say that winc(p) = 0.) By

the analysis of Stage 2c, we know that the worker will only be retained if Sι ≥ Sι. Two

cases require some analysis, because they involve renegotiation.

Case (i) (Paying to retain): Sι > Sι > winc. In this case, the employed specialist is

worth more at the incumbent firm than outside, but will not stay in the firm under the

current contract. In making a new offer, the employer has two main alternatives: Either

pay just enough for the specialist to stay, but shade on effort, or pay enough for the

specialist to stay and not shade. Recall that Tι = Sι − Sι. If the incumbent employer

offers a new level of pay w′ ≥ Sι + βTι, the specialist will stay and exert effort eN = 1.

Since effort is constant for these offers, and they suffice to beat outside competition, the

best such offer is wjs = Sι + βTι, yielding a profit (1 − β)Tι. If the incumbent employer

instead offers w′ ∈[Sι, Sι + βTι

), the specialist will stay and exert effort e. Since the

effort level is insensitive to the exact offer in this interval, we only need to consider the

offer w′ = Sι. This offer yields a profit for the firm of eSι− Sι. Comparing the two profits,

it is a straightforward computation to see that

(1− β)Tι > eSι − Sι

if and only if

β <Sι

Sι − Sι(1− e).

This condition always holds because Sι ≥ 0, Sι ≥ 0, Sι− Sι > 0, (hence Sι/(Sι− Sι) > 1),

and β < 1 − e (by Assumption 2). Hence we have established that in this case, the

equilibrium (net wage) renegotiation offer by the incumbent employer is

wr = Sι + β(Sι − Sι), (15)

where we let the subscript r indicate that this is the renegotiated wage.

Case (ii) (Paying to depart): winc > Sι > Sι. In this case, the specialist is worth more

on the outside, but agreed pay is higher than the best outside offer. There is a range of

mutually profitable contracts, where the incumbent employer increases the pay in case

of departure by at least winc − Sι. Indeed, the equilibrium (increase in) severance pay is

exactly this amount. Even if all the surplus from renegotiation goes to the employer, the

specialist will not remain in the firm. Therefore, shading is not a concern in this case.

In the remaining cases, there is no renegotiation. Either, the incumbent’s contract

15

is generous enough to retain the specialist or the outside option Sι is so attractive that

there is no profitable offer that can make the specialist stay.

We are now ready to characterize the equilibrium Stage 1 contracts, for each of the

different contractual forms.

4.3 Unconstrained contracts

Suppose firms can make unconstrained contract proposals. Then, the unique equilibrium

outcome gives each specialist a fixed salary corresponding to the specialist’s expected

value conditional on an optimal assignment.

Proposition 2 When contracts are unconstrained, a fully optimal outcome is attained

in an equilibrium where each firm in industry i∗1 offers talented workers contracts of the

form

wu =

S∗ if i2 = i∗1;

S∗ − Sι if i2 6= i∗1.(16)

There is no equilibrium in which any talented worker gets any other final pay.

Proof. See Appendix.

The idea is simple. The employee gets a fixed wage regardless of what task is conducted

or where, but has to pay the original employer the full value of any surplus generated for

another employer.

These ideal contracts are unrealistic. For example, they require that the firm is

able to force the worker to pay potentially large penalties in case of departure. The

contracts that we consider in the rest of the paper all share the realistic assumption that

firms cannot penalize their employees for leaving. With this constraint, in period 2 an

employed specialist must now be paid, in each state, at least the surplus that is attainable

through re-matching with some other firm, Sι.

4.4 Spot contracts

Under a spot contract, there is no commitment on either side. For this reason, expected

gains from subsequent contracting must be distributed immediately. The spot contract

thus specifies a fixed compensation ws that is contingent merely on training with a firm.

In the second period, the worker is offered a new contract in a competition between all

firms. The equilibrium outcome for a worker can be described as follows.

16

Proposition 3 Under spot contracting, the unique equilibrium outcome entails, for each

talented worker, (i) an optimal second-period assignment i∗2(ι) and (ii) the remuneration

rs = ws + Sι + max{

0, βTι},

where

ws = E[max

{0, (1− β)Tι

}].

Moreover,

i1 ∈ arg maxι∈I

E [u(rs)] .

Proof. See Appendix.

The interpretation runs as follows. In period 2, the worker always obtains at least

the best outside option Sι; this is the second term in the expression for rs. In addition,

when it is optimal to stay with the incumbent employer, the worker obtains a fraction β

of the net surplus; this is the third term. The expected value of the remaining fraction

is paid out as a lump-sum, i.e., in all states; this is the first term. The expression for the

wage also has a simple interpretation. It is the expected period 2 rent that accrues to

the employer because the worker has acquired firm-specific capital.

Corollary 1 Spot contracting always entails inefficient risk-sharing.

Since the pay always varies with the best outside option – whether that alternative

employment is in the same industry or in an other industry – the worker’s pay is always

uncertain. As a result, the worker will care about industry volatility as well as about

expected pay when choosing occupation.

4.5 Salary contracts

There are two important differences between a salary contract and a spot contract. First,

unlike the spot contract which involves a sign-on payment irrespective of whether the

worker subsequently stays or leaves, an optimal salary contract does not pay out anything

in case the worker leaves for a better offer. Second, the salary pins down pay in all states

where the worker’s outside option is not binding; the worker can only renegotiate when

there is a credible threat of departing. For both reasons, there will typically be relatively

attractive states p in which the salary compensation is lower than under spot contracts.

In return, the worker is paid more in the relatively bad states.

Proposition 4 Under salary contracting, the unique equilibrium outcome entails, for

each talented worker, (i) an optimal second-period assignment i∗2(ι) and (ii) the remuner-

17

ation

ra = max{wa, Sι, Sι + βTι

},

where wa is the largest solution to

E[ra(w)] = E[Sι].

Proof. See Appendix.

In this case, we refrain from writing out the explicit formula for the equilibrium wage

wa. Essentially, the wage equalizes the expected rents that the initial employer harvests

under favorable realizations of p (because of the employees firm-specific human capital)

and the expected losses that the initial employer incurs under unfavorable realizations,

thereby ensuring that no employer earns any rent.

Comparing ra to rs, we can show that risk-averse workers are generally better off

under salary contracts than under spot contracts.

Proposition 5 Ex ante, for any talented worker and any chosen industry, the equilib-

rium remuneration under salary contract competition second-order stochastically domi-

nates the equilibrium remuneration under spot contract competition.

Proof. See Appendix.

A salary contract can even be first-best. That happens under the following condition.

Corollary 2 A salary contract is optimal if and only if

S∗ ≥ maxi∈I

(1 + 1ικθ)(pi − wg)ai. (17)

where 1ι is an indicator function taking value 1 if the specialist is assigned to the same

firm in both periods and 0 otherwise.

The proof is straightforward from inspection of Proposition 4: If (17) holds, worker j will

never get a wage offer exceeding the flat salary, that is wa = S∗ > Sι(p) for all feasible p.

It is efficient to retain the worker if and only if

(pι − wg)(1 + κ)aι ≥ maxi 6=ι

(pi − wg)ai. (18)

If inequality (18) does not hold, the current employer optimally offers severance pay

S∗ − maxi 6=ι(pi − wg)ai and the new firm offers the wage maxi 6=ι(pi − wg)ai, which in

combination yield the same fixed salary as if the worker stayed. The worker thus receives

a remuneration that is independent of p and is efficiently matched. That is, there is no

18

need for variable pay if the worker’s productivity in other firms or industries never veer

above the expected productivity of the worker in the initial job.

4.6 Simple performance contracts

Simple performance contracts allow employers to tie pay to the performance of the own

firm, but not explicitly to the performance of other firms.20 We think of this as the

contract environment that comes closest to the contracting environment that faces large

firms in modern economies.

Observe that we do not impose any functional form on this contract. Nonetheless,

equilibrium contracts turn out to be piece-wise linear.

4.6.1 Narrow specialization

As a starting point let us assume that each worker is talented in at most one industry,

that is, aj has at most one positive component. Call this narrow specialization. Under

narrow specialization, the inability to condition contracts on subsequent offers from other

industries obviously does not matter.

Proposition 6 If workers are narrowly specialized, any competitive equilibrium outcome

is attained through simple performance contracts of the form

wc(pι) = max{wc, (pι − wg)aι(1 + θκ)}. (19)

where the wage floor, wc, is the the highest solution to

E[wc(pι)] = E[Sι].

Proof. Any alternative break-even contract to wc(pι) will increase the risk exposure of

the worker and keep the expected wage constant: Suppose the worker is paid less than

wc(pι) in cases where wc(pι) = (pι−wg)aι(1+θκ), the worker will renegotiate the contract

and thereby reduce the profit of the firm which will induce the firm to reduce the wage

floor and thereby increase the risk exposure of the firm. Suppose wc(pι) = wc, reduced

pay will increase the profit of the firm and not be a competitive outcome. Suppose the

worker is paid more than wc(pι) in cases where wc(pι) = (pι−wg)aι(1 + θκ). The worker

will be paid strictly more his outside offer and the firm will need to reduce the wage floor

20In the present setting, tying pay to performance of the own firm is equivalent to tying it to the per-formance of the own industry. Below, we comment on the case in which there are additional idiosyncraticshocks at the firm level.

19

in order to keep the break-even constraint. In both cases the worker will be exposed to

more risk, receive the same expected wage, and prefer contract wc(pι) instead.

The contract ensures that outside offers from within the industry are not attractive,

and thus prevents renegotiation. This is valuable, because renegotiation would entail pay

that is strictly above the outside offer, and thus increase the ex ante variance.

Figure 1 displays the simple performance contract (for simplicity, we have set wg = 0).

The slope b in the figure is aι(1+θκ), illustrating that more talented workers and workers

with more portable human capital have steeper performance pay schedules.

wc(pι)

pιwc/b

wc

bpι

Figure 1: A simple performance contract

A fairly immediate corollary is that there is more performance pay when uncertainty

is larger.

Corollary 3 Assume narrow specialization. Let h(pι) be a mean-preserving spread of

h(pι). The the wage floor, wc is weakly lower and the the expected performance pay is

weakly larger under h(pι) than under h(pι). The relationship is strict if∫ p

0

H(pι)dpι >

∫ p

0

H(pι)dpι,

where p = wc/(1 + θκ)aι.

20

Proof. To be added.

In this sense, the model makes the opposite prediction of the classical incentive contract

model, according to which there should be less performance pay when uncertainty is

greater. Our model thus provides a new reason for why the link between risk and incen-

tives is tenuous in the data; see Prendergast (2002) for an overview of relevant evidence

(as well as another explanation).

4.6.2 Multiple talents

If a worker has multiple talents, the worker should move across industries whenever some

alternative industry offers higher productivity. Under the following additional assumption

about the distribution of states p, the simple performance contact will take the same

piece-wise linear shape as described in Proposition 6 .

Assumption 3 Define z = (1 + κ)/(1 + θκ) and m = max{Si(p)}i 6=ι. The associated

distribution, g(m(p)), satisfies (i) G(m)/g(m) is increasing for m ≥ wc(pι) (a monotone

hazard rate condition), and (ii)

G(wc(pι))

wc(pι) g(wc(pι))> z − 1. (20)

This is a mild assumption.21

The role of Assumption 3 is to ensure that it is not optimal to pay above the intra-

industry offer in order to implement an analogous reduction of the expected cost of fending

off inter-industry offers.

Proposition 7 Under competition in simple performance contracts, the unique equilib-

rium outcome entails, for each talented worker, (i) an optimal second-period assignment

i∗2(ι) and (ii) the remuneration

rc = max{wc, Sι + 1ιβTι

}, (21)

where 1ι = 1 if Sι > Sι > Sι and 0 otherwise, and wc is the unique solution to

E[rc(w)] = E[Sι]. (22)

21Note that (i) holds for any single-peaked distribution with a peak below wc(pι). To interpret (ii),note that G(wc(pι)) is the probability that the employee is better paid in the original job than themost attractive potential offer from a different industry. To interpret g(wc(pι)), consider for referencea uniform distribution on the interval [0, x]. This distribution has constant density g = 1/x. For anyx > wc(pι), the left-hand side of (20) would then be above G(wc(pι)). Thus, for any function on the sameinterval that has less than average density at points m ≥ wc(pι), it is sufficient that G(wc(pι)) > z − 1.

21

(iii) Equilibrium wage contracts are given by wc(pι) with a wage floor, wc, determined by

equation (22).

Proof. The proof is analogous to the proof of Proposition 4. There are only two dif-

ferences. First, the simple performance contract is not renegotiated in the case that the

best outside offer comes from industry ι. Second, we need to prove that it is not opti-

mal to have a performance-pay component exceeding the best intra-industry offer (see

Appendix).

Observe that multiple talents and attractive offers from alternative industries can

lead to renegotiations and higher pay in states where the original employer offer the best

match. Frequent wage-renegotiations will drive down the wage floor and increase the risk

exposure of the worker. On the other hand, if outside offers lead to efficient turnover,

workers productivity will increase and wages will improve. The negative effects of inter-

industry competition are important when workers are very risk averse and outside offers

seldom implies efficient turnovers while the positive effects are more important when

outside offers leads to efficient turnover and higher productivity of workers.

Figure 2 illustrates the various final outcomes for the special case of two industries,

with ι = 1.

S2(p2)

S1(p1)

wc

S1(wc/b)

bp1

S1(p1) = S2(p2)

Contract implemented

Pay renegotiated

S2(p2) + βT1

wc bp1

Turnover

S2(p2)

Severancepay

Figure 2: Pay and turnover as functions of p = (p1, p2), with ι = 1.

22

Recall that there are two differences between the contract wc and the actual pay rc.

One is that the worker sometimes optimally moves to another industry which pays more.

The other is that the worker’s best outside option comes from such a different industry,

yet is worth less than the worker’s human capital in the initial firm. In the latter case,

the wage is renegotiated upward to a level that is consistent with the worker staying and

not shirking.

The typical case of turnover occurs when the own industry performs relatively poorly;

pι is low. This is consistent with the central regularity emphasized by Jenter and Kanaan

(2015); CEOs are mostly fired after bad firm performance caused by factors beyond the

manager’s control, especially when the firm’s industry is performing poorly.22

As in the case of a pure salary contract, the fact that the worker is being promised

a pay of at least wc by the original employer does not preclude the possibility that the

worker leaves for an outside offer that is less than wc. Whenever Sι < Sι < wc, it is

desirable that the worker leaves. Since the worker cannot be commanded to leave, the

employer will offer a severance pay of wc−Sι to make up for the lost wage; this transaction

yields a gain of Sι − Sι to the employer; the worker gains nothing, as the severance pay

and the pay from the new employer add up to wc. Since the wage floor is higher under

competition in simple performance contracts than under competition in salary contracts,

the severance pay will be higher under simple performance contracts.

To summarize, the model is consistent with the evidence that CEO severance pay

is usually awarded on a discretionary basis by the board of directors and not according

to terms of an employment agreement, was found by Yermack (2006).23 Likewise, the

feature that severance pay makes up for the loss in expected compensation rhymes well

with Yermack’s interpretation of the severance pay data: “boards use severance pay to

assure CEOs of a minimum lifetime wage level.”

Before interpreting this result further, we note that simple performance contracts

dominate salary contracts whenever a salary contract is not strictly optimal.

Proposition 8 Suppose (17) fails. Ex ante, for any talented worker, the remuneration

under the equilibrium simple performance contract second-order stochastically dominates

the remuneration under the equilibrium salary contract.

Proof. See Appendix.

22As Jenter and Kanaan note, this behavior by corporate boards is inexplicable, or suggestive ofirrationality, in the incentive provision framework. Once we consider the retention motive, however, itmakes perfect sense to keep talented workers when the industry is profitable and growing and releasethem to other more productive jobs elsewhere when the industry declines.

23Discretionary severance pay is difficult to reconcile with models that emphasize ex ante incentiveissues, such as those of Almazan and Suarez (2003), Inderst and Mueller(2010), and Manso (2011). Inthese models, severance pay is only useful if contracted in advance.

23

The idea is simple. There is enough uncertainty that any equilibrium fixed salary con-

tract would occasionally have to be renegotiated. The worker’s pay is higher under such

renegotiation than it would have been under the simple performance contract. Hence, the

base wage is higher under a simple performance contract, which is therefore preferable.

4.7 Additional sources of variation

So far, we have assumed that all uncertainty is industry-specific. What if there is also

firm-specific uncertainty? Suppose pfi = εfpi, where εf is a firm-specific shock with mean

1. The most attractive outside offer will now come from the firm with the most favorable

realization of the firm-specific shock. We have assumed that it is impossible to make the

contract dependent on the stock price of a particular other firm. However, suppose it

is possible to let the contract depend on the industry stock-price index, which in turn

would depend on the average state of industry ι, pavgι . Suppose moreover that there are

many firms in the industry.

Suppose that firm-specific uncertainty is small, say, with εf ∈ [1− δ/2, 1+ δ/2]. Then,

the worker ought not to leave the original employer. If a firm uses own performance to

construct the simple performance contract, it has to offer the following contract to surely

fend of any offers from firms in the same industry:

(1 + δ) pι (1 + θκ).

Note that firm-specific uncertainty in addition to industry-specific uncertainty implies

more use of variable pay. If the firm instead offers a contract based on average performance

(and the number of firms is very large), it suffices to offer

(1 +1

2δ) pavg (1 + θκ),

which involves less performance wage and a higher wage floor and will retain workers

in the same cases as above. However, such index options are only preferable as long

as the industry index is a better indicator of the outside option than is the firm’s own

stock price.24 For example, suppose that the portability θ varies across firms within the

same industry and is higher between firms that have more correlated stock prices. Then,

the ideal contract would have been to condition pay on the stock price of the closest

competitors. But since that is not possible, it could be better to base performance-

related pay on the own firm’s stock price than on a broad index of all the firms in the

24Observe that pay depends positively on the industry index rather than negatively; negative depen-dence is the recommendation when the purpose is to reward high effort.

24

industry.25 At any rate, when firms’ stock prices are closely correlated with the industry

average, any loss from conditioning on the own firm’s stock price rather than the index

will be small.

Another natural extension of the model is to allow match-specific shocks. Suppose the

employee suddenly gets to dislike the current firm or fancy another. If these shocks are

large enough, turnover is motivated in equilibrium. The notable additional insight is that

a given level of match-specific shock is more likely to produce turnover if human capital

is more portable. Therefore, we should expect to see high turnover to go hand-in-hand

with high levels of “pay for luck.”

5 Interpretation and Implications

Why is the simple performance contract piece-wise linear in equilibrium? Formally, the

reason is plain. The fraction θ of the industry-specific skill that is portable is a constant.

Therefore the value of portable human capital is p1κθa1. If instead θ had been dependent

on the state p, the optimal contract would have been non-linear.26 In our view, it is

difficult to argue that there is a direct technological impact of p on θ.

5.1 Stock options

In reality, a typical piece-wise compensation contract involves stock options. As is easily

seen, a standard call option admits implementation of the equilibrium contract.

Corollary 4 Simple performance contracts can be implemented through a combination

of a salary and an option to purchase the own firm’s stock.

Specifically, if we interpret pι as the stock price, the strike price of worker j’s stock

option, psι solves the equation wc = pι(1 + κθ)aι. That is, the stock option becomes

valuable exactly at the point where outside offers exceed the worker’s salary, wc. In order

to prevent (inefficient) turnover, the size of the option package must be proportional to

the worker’s portable human capital, κθaι.

5.2 The role of portability

Preventing people from leaving with valuable human capital is a central element of good

corporate governance; for detailed arguments, see for example Rajan and Zingales (2001)

25In reality, a related problem with using industry indexes is that they do not capture the competitionfrom new entrants.

26Ellingsen and Kristiansen (2011) advance a similar argument for the optimality of simple securities.

25

and references therein. Which are the factors affecting portability, and how do they show

up in empirical work?

Portability is affected by two main factors. Portability is higher (i) when the initial

employer has weak property rights over the employee’s human capital (and it is difficult

to transfer such rights from the employee to the employer), and (ii) when there exist other

firms that can make good use of the human capital. With these two factors in mind, we

see that the positive association between portability and variable pay resonates well with

many empirical observations:

1. “Knowledge” firms utilize stocks and especially stock options to a larger degree

than do “brick and mortar” firms; see Anderson et al. (2000); Ittner et al. (2003);

Murphy (2003); Oyer and Schaefer (2005)). The knowledge firms themselves report

that performance-based pay is primarily used for retention purposes; see Ittner et al.

(2003)). We think of knowledge firms as having high θκ relative to brick-and-mortar

firms.

2. Option-based compensation is particularly common in “growth firms”, both for ex-

ecutives (e.g., Smith and Watts (1992); Gaver and Gaver (1993); Mehran (1995);

Himmelberg et al. (1999); Palia (2001)) and non-executives (e.g., Core and Guay

(2001)). We think of growth firms as being more recent and having less well pro-

tected technologies, hence a higher portability θ.

3. Industries with a higher fraction of outside executives have both a larger fraction of

performance related pay and a smaller degree of indexing, i.e., more pay for luck;

see Cremers and Grinstein (2014); also Murphy and Zabojnik (2004, 2006). For

reasons provided above, we think of high observed mobility as indicating high θ.

4. Managers have weaker performance incentives in regulated sectors; see Murphy(1999);

Frydman and Saks (2010)). Regulated sectors are often operated by a monopoly,

and generally have fewer potential alternative employers. Hence, θ is low.

5. There is less performance-based pay in family firms (e.g., Kole (1997); Anderson

and Reeb (2003); Bandiera et al (2010)), especially when the manager is a family

member (Gomez-Mejia et al. (2003)). We think that family members are more

reluctant to move, hence θ is low.

6. There is less mobility of managers, and managers have weaker performance incen-

tives in jurisdictions with stronger enforcement of noncompete employment clauses

26

(Garmaise, 2011). This is perhaps the most direct interpretation of our theory – θ

is smaller when noncompete clauses are stronger.27

7. There is a weaker link between pay and industry-wide shocks (“luck”) in companies

with large owners, especially when these large owners sit on the company’s board

of directors (Bertrand and Mullainathan (2001); see also Fahlenbrach (2009)). We

think of large owners as exercising better control and hence reducing θ.28

8. Performance hurdles for option contracts are increasing in the quality of corporate

governance; see Bettis et al. (2010). This point is similar to the previous one. If

good corporate governance entails better safeguards against workers departing with

valuable assets, good corporate governance will be associated with a higher hurdle

price psι (see section 5.1).

In addition to this cross-section evidence, Murphy and Zabojnik (2004, 2006) argue that

the relative importance of transferable talent has increased over time, as evidenced by the

executives’ education as well as the increasing frequency of externally hired executives.

If this view is accepted, our model can account for the increase in variable pay over the

last few decades that is documented by Frydman and Saks (2010).

5.3 Hedging

With a simple performance contract, the employee carries all the risk associated with the

call option package. Suppose now that the worker tries to hedge this risk by selling an

identical set of call options for a fixed price in a perfect financial market. As we shall

now see, such a contract combines desirable insurance with optimal bonding.

Consider a state p in which the options are “in the money”; they are worth m(pι) =

bpι − wc > 0. Then, in case the worker stays, she may simply exercise her own options

in order to satisfy the call of the buyer of the options. On the other hand, if the worker

departs, she forfeits the call options promised by the employer. Thus, to satisfy the

buyer’s call, she has to pay m(pι) from her own pocket. Thus, she will depart if and only

if Sι − bpι > 0. In other words, there is departure if and only if it is optimal that the

worker is employed in another industry. The hedging does not entail any allocation loss,

and it eliminates all the remuneration risk for states p such that the worker is optimally

27A related finding is that mobility is smaller when firms defend their patents more aggressively, seeGanco, Ziedonis, and Agarwal (2015).

28A problem with this argument is that the presence of large owners could be endogenous; they arethere because θ is high to begin with; a challenge for empirical work would be to tease out exogenousvariation in ownership.

27

allocated to the initial industry. In case retention is sure to be ex post optimal, the

first-best is thus attained.

Proposition 9 Suppose the financial market is perfect and that workers are narrowly

specialized. Then a simple performance contract wc(pι) that can be freely hedged imple-

ments the first-best outcome.

The model thus offers both a straightforward normative argument for allowing hedging

and a positive argument for why hedging occurs.29

6 Final remarks

Critics of executive compensation practices often argue that non-indexed performance

pay, high pay-floors, and discretionary severance pay cannot be efficient, and are thus

prima facie evidence of managerial rent-extraction. To the contrary, we find that optimal

contracts have all three features; in fact, these features could be socially desirable even

as instruments to reduce pay inequality. In our model, a commitment to non-indexed

performance pay serves primarily to reduce the very highest realizations of pay that would

be observed under fixed-salary contracting.

Our model highlights the importance of the retention motive for the shape of com-

pensation contracts: When the worker’s outside option never binds, a fixed salary is the

equilibrium outcome, but when market conditions are sufficiently variable, compensation

contracts will be designed to match the worker’s most attractive outside employment op-

portunity. While it is possible to adapt to outside offers after the fact, such renegotiation

does not mimic an optimal ex ante contract. Pay variability is lower when variable pay

components are contracted ex ante than when they are negotiated ex post.

Deviations from plain salary contracting will be especially large in volatile industries

where industry-specific human capital is portable across firms. If our model is right,

such industries are likely to continue their “controversial compensation practices,” such

as granting stock options that pay lavishly when the industry booms and provide discre-

tionary severance pay when it busts. To the extent that new public or corporate policies

– such as higher taxation of stock options or more indexation of contracts – are put in

place in order to curb such pay practices, the total cost of compensating the most talented

employees could be going up rather than down.

29Among the CEO contracts studied by Schwab and Thomas (2006), only five out of 271 restrictedhedging. For other studies of the prevalence of hedging, see Bettis, Bizjak, and Lemmon (2001) andKallunki, Nilsson, and Hellstrom (2009).

28

References

Almazan, A. and Suarez, J. (2003), Entrenchment and severance pay in optimal gover-

nance structures, Journal of Finance 58, 519–547.

Anderson, M. C., Banker, R. D. and Ravindran, S. (2000), Executive compensation in

the information technology industry, Management Science 46, 530–547.

Anderson, R. C. and Reeb, D. (2003), Founding-family ownership and firm performance:

Evidence from S & P 500, Journal of Finance 58, 1301–1328.

Bandiera, O., Guiso, L., Prat, A. and Sadun, R. (2010), Matching firms, managers and

incentives, Cepr working paper 7207 (Forthcoming Journal of Labor Economics).

Bartling, B. and Schmidt, K.M. (2015), Reference points, social norms, and fairness in

contract renegotiations, Journal of the European Economic Association 13, 98-129.

Bebchuk, L.A. and Fried, J.M. (2004), Pay without performance: The unfulfilled promise

of excecutive compensation Harvard University Press.

Bebchuk, L.A, Fried, J.M and Walker, D.I, (2002), Managerial power and rent extraction

in the design of executive compensation, University of Chicago Law Review 69, 751–846.

Bertrand, M. and Mullainathan, S. (2001), Are CEOs rewarded for luck? The ones

without principals are, Quarterly Journal of Economics 116, 901–932.

Bell, B. and van Reenen, J. (2016), CEO Pay and the Rise of Relative Performance

Contracts: A Question of Governance, CEP Discussion Paper No. 1439.

Bettis, C., Bizjak, J., Coles, J. and Kalpathy, S. (2010), Stock and option grants with

performance-based vesting provisions, Review of Financial Studies 23, 3849–3888.

Blakemore, A. E., Low, S. A. and Ormiston, M. B. (1987), Employment bonuses and

labor turnover, Journal of Labor Economics 5, S114–S135.

Brandts, J., Charness, G. and Ellman, M. (2016), Let’s talk: How communication affects

contract design, Journal of the European Economic Association forthcoming.

Core, J. E. and Guay, W. R. (2001), Stock option plans for nonexecutive employees,

Journal of Financial Economics 61, 253–87.

Cremers, M. and Grinstein, Y. (2014), Does the market for CEO talent explain contro-

versial CEO pay practices? Review of Finance 18, 921–960.

29

Dutta, S. (2003), Capital budgeting and managerial compensation: Incentive and reten-

tion effects, Accounting Review 78, 71–93.

Edmans, A., Gabaix, X. and Landier, A. (2009), A multiplicative model of optimal CEO

incentives in market equilibrium, Review of Financial Studies 22, 4881–4917.

Ellingsen, T. and Kristiansen, E. G. (2011), Financial contracting under imperfect en-

forcement, Quarterly Journal of Economics 126, 323–371.

Fahlenbrach, R. (2009), Shareholder rights, boards, and executive compensation, Journal

of Political Economy 13, 81–113.

Fehr, E., Hart, O. and Zehnder, C. (2009), Contracts, reference points, and competition–

behavioral effects of the fundamental transformation. Journal of the European Economic

Association 7 (2-3), 561–572.

Fehr, E., Hart, O. and Zehnder, C. (2011), Contracts as reference points–experimental

evidence, American Economic Review 101(2): 493–525.

Fehr, E., Hart, O. and Zehnder, C. (2015), How do informal agreements and revision

shape contractual reference points? Journal of the European Economic Association 13

(1), 1–28.

Fehr, E. and List, J. (2004), The hidden costs and returns of incentives – trust and

trustworthiness among CEOs, Journal of the European Economic Association 2(5), 743-

771.

Frydman, C. and Saks, R. (2010), Executive compensation: A new view from a long-term

perspective, 1936 – 2005, Review of Financial Studies 23, 2099 – 2138.

Fudenberg, D., Holmstrom, B. and Milgrom, P. (1990), Short-term contracts and long-

term agency relationships, Journal of Economic Theory 51(1), 1-31.

Gabaix, X. and Landier, A. (2008), Why has CEO pay increased so much? Quarterly

Journal of Economics 123(1), 49–100.

Ganco, M., Ziedonis, R. and Agarwal, R. (2015), More stars stay, but the brightest ones

still leave: Job hopping in the shadow of patent enforcement, Strategic Management

Journal 36, 659–685.

Garmaise, M.J. (2011), Ties that truly bind: Non-competition agreements, executive

compensation and firm investment, Journal of Law, Economics, and Organization, 27,

376-425.

30

Garvey, G. and Milbourn, T. T. (2006), Asymmetric benchmarking in compensation:

Executives are rewarded for good luck but not penalized for bad, Journal of Financial

Economics 82, 197–225.

Gaver, J. J. and Gaver, K. M. (1993), Additional evidence on the association between the

investment opportunity set and corporate financing, dividend, and compensation policies,

Journal of Accounting and Economics 16, 125–160.

Giannetti, M. (2011), Serial CEO incentives and the structure of managerial contracts,

Journal of Financial Intermediation 20, 633–662.

Gomez-Mejia, L. R., Larraza-Kintana, M. and Makir, M. (2003), The determinants of ex-

ecutive compensation in family-controlled public corporations, Academy of Management

Journal 46, 226–237.

Groysberg, B., Lee, L.-E. and Nanda, A. (2008), Can they take it with them? The

portability of star knowledge workers’ performance, Management Science 54, 1213–1230.

Harris, M. and Holmstrom, B. (1982), A theory of wage dynamics, Review of Economic

Studies 49, 315 – 333.

Hart, O. (2009), Hold-up, asset ownership, and reference points, Quarterly Journal of

Economics 124(1), 267–300.

Hart, O. and Holmstrom, B.(2010), A theory of firm scope, Quarterly Journal of Eco-

nomics 125(2), 483–513.

Hart, O. and Moore, J. (2008), Contracts as reference points, Quarterly Journal of Eco-

nomics 123(1), 1–48.

Hashimoto, M. (1979), Payments, on-the-job training, and lifetime employment in Japan,

Journal of Political Economy 87, 1086–1104.

Hermalin, B. E. and Weisbach, M. (1998), Endogenously chosen boards of directors and

their monitoring of the CEO, American Economic Review 88, 96–118.

Himmelberg, C. P., Hubbard, R. G. and Palia, D. (1999), Understanding the determinants

of managerial ownership and the link between ownership and performance, Journal of

Financial Economics 53, 353–384.

Holmstrom, B. (1979), Moral hazard and observability, Bell Journal of Economics 10,

74–91.

31

Holmstrom, B. and Milgrom, P. (1987), Aggregation and linearity in the provision of

intertemporal incentives, Econometrica 55, 303–328.

Holmstrom, B. and Milgrom, P. (1991),

Holmstrom, B. and Ricart i Costa J. (1986), Managerial incentives and capital manage-

ment, Quarterly Journal of Economics 101, 835–560.

Holmstrom, B. and Kaplan, S. (2003), The state of US corporate governance, Journal of

Applied Corporate Finance 15, 8–20.

Hubbard, R. G. (2005), Pay without performance: a market equilibrium critique, The

Journal of Corporate Law 30, 717–720.

Inderst, R. and Mueller, H. M. (2010), CEO replacement under private information,

Review of Financial Studies 23, 2935–2969.

Ittner, C. D., Lambert, R. A. and Larcker, D. F. (2003), The structure and performance

consequences of equity grants to employees of new economy firms, Journal of Accounting

and Economics 34, 89–127.

Jenter, D. and Kanaan, F. (2015), CEO turnover and relative performance evaluation,

Journal of Finance 70, 2155–2184.

Johnson, S. A. and Tian, Y. S. (2000), The value and incentive effects of nontraditional

executive stock option plans, Journal of Financial Economics 57, 3–34.

Kallunki, J. P., Nilsson, H. and Hellstrom, J. (2009), Why do insiders trade? Evidence

based on unique data on Swedish insiders, Journal of Accounting and Economics 48,

37–53.

Kole, S. R. (1997), The complexity of compensation contracts, Journal of Financial

Economics 43, 79–104.

Kvaløy, O. and Olsen, T. (2012), The rise of individual performance pay, Journal of

Economics and Management Strategy 21, 493–518.

Lazear, E. P. and Oyer, P. (2012), Personnel economics, in R. Gibbons and D. J. Roberts,

eds., Handbook of Organizational Economics vol. 3, New York: Elsevier.

Manso, G. (2011), Motivating innovation, Review of Financial Studies 66, 1823–1869.

Marx, M., Strumsky, D. and Fleming, L. (2009), Mobility, skill, and the Michigan non-

compete experiment, Management Science 55, 875–889.

32

Mehran, H. (1995), Executive compensation structure, ownership, and firm performance,

Journal of Financial Economics 38, 163–184.

Milbourn, T. T. (2003), CEO reputation and stock-based compensation, Journal of Fi-

nancial Economics 68, 233–262.

Murphy, K. and Zabojnik, J. (2004), CEO pay and appointments: A market-based ex-

planation for recent trends, American Economic Review, Papers and Proceedings 94,

192–196.

Murphy, K. and Zabojnik, J. (2006), Managerial capital and the market for CEOs,

Queen’s economics department working paper no. 1110.

Murphy, K. J. (1999), Executive compensation, in A. Orley and D. Card, eds., Handbook

of Labor Economics vol. 3, New York: Elsevier.

Murphy, K. J. (2003), Stock-based pay in new economy firms, Journal of Accounting and

Economics 34, 129–147.

Oyer, P. (2004), Why do firms use incentives that have no incentive effects?, Journal of

Finance 59, 1619–1649.

Oyer, P. and Schaefer, S. (2005), Why do some firms give stock options to all employees?:

An empirical examination of alternative theories, Journal of Financial Economics 76,

99–133.

Pakes, A. and Nitzan, S. (1983), Optimal contracts for research personnel, research em-

ployment, and the establishment of ”rival” enterprises, Journal of Labor Economics 1,

345–365.

Palia, D. (2001), The endogeneity of managerial compensation in firm valuation: A

solution, Review of Financial Studies 14, 735–764.

Prendergast, C. (2002), The tenuous trade-off between risk and incentives, Journal of

Political Economy 110, 1071–1102.

Rajan, R. and Zingales, L. (2001), The firm as a dedicated hierarchy: A theory of the

origins and growth of firms, Quarterly Journal of Economics 116, 805–851.

Rajgopal, S., Shevlin, T. and Zamora, V. (2006), CEOs’ outside employment opportuni-

ties and the lack of relative performance evaluation in compensation contracts, Journal

of Finance 61, 1813–1844.

33

Rosen, S. (1992), Contracts and the market for executives, in: L. Werin and H. Wijkander

(eds.), Contract Economics. Cambridge, MA and Oxford: Blackwell, 181–211.

Schwab, S.J. and Thomas, R.S. (2006), An empirical analysis of CEO employment con-

tracts: What do top executives bargain for? Washington and Lee Law Review 65(1),

231-270.

Smith, C. J. and Watts, R. L. (1992), The investment opportunity set and corporate

financing, dividend, and compensation policies, Journal of Financial Economics 32, 263–

292.

Tervio, M. (2008), The difference that CEOs make: An assignment model approach,

American Economic Review 98(3), 642–668.

Yermack, D. (1995), Do corporations award CEO stock options effectively?, Journal of

Financial Economics 39, 237–269.

Yermack, D. (2006), Golden handshakes: Separation pay for retired and dismissed CEOs,

Journal of Accounting and Economics 41, 237–256.

Appendix: Proofs

Proof of Proposition 1

The proof is by contradiction. If (ii) is violated, then there is some constant remuneration

rc < E[r] such that the risk-averse worker would be better off with rc than with r. And

any one of the risk-neutral firms are willing to trade the fixed payment rc in return

for the stochastic r. Consider a violation of the assignment rule (i). Then, given (ii),

a reassignment of the worker to the payoff-maximizing industry (in periods 1 and 2

respectively) would generate an increase in output, which could be shared among the

original employer and the new employer.

Proof of Proposition 2

We first show that this contracting outcome is part of a subgame-perfect equilibrium.

Note that the worker pays a penalty equal to the maximum surplus Sι if the worker

departs. From the analysis of Stage 2c, we know that workers will earn Sι if in period

2 they move to another employer than the incumbent. If Sι ≥ Sι, so the incumbent

employer is the optimal employer, it thus does not pay to depart. If instead Sι < Sι, i.e.,

the original employer is not the optimal employer in period 2, then the final net pay is S∗

34

regardless of whether the worker stays or leaves. Let the worker resolve the indifference

according to the surplus maximizing assignment. Finally, note that no firm can gain by

making another offer. If the offer is more generous, it creates a loss. If it is less generous,

worker j signs another contract.

We finally show that no subgame-perfect equilibria support any other final pay. (i)

Contracts that yield an expected pay is above S∗ are clearly loss-making for a risk-neutral

employer. (ii) Suppose there were an equilibrium in which a talented worker is employed

at some contract w yielding an expected total pay strictly below S∗. Suppose first that

a departing worker is underpaid by a new employer. This cannot be consistent with

equilibrium, because the new employer then earns a positive rent; it would be profitable

for another employer in the same industry to bid higher. Underpayment must thus be

associated with states in which the worker is retained. But if so, the worker is underpaid

on average by the initial employer. But this cannot be consistent with equilibrium either.

Now, the incumbent employer earns a rent, and it would have been profitable for another

employer in this industry to bid higher.

Proof of Proposition 3

Start in period 2, after realization of the state p. If Sι < Sι, we know from the analysis

of Stages 2c-d that any SPE has the feature that the worker leaves and is paid Sι by the

new employer.

If Sι > Sι, we know from the analysis of Stage 2b, that the renegotiated wage is given

by equation (15).

Next, since there is Bertrand competition between risk-neutral employers, the fixed

pay ws must be such that the worker’s expected pay equals the expected value of the

worker’s services, E[rs(w)] = E[Sι]. This means that the salary must equal the expected

benefit that the employer will reap in period 2, which is E[max

{0, (1− β)Tι

}]≥ 0.

Finally, the characterization of the initial industry choice, i1 is self-evident: A worker

j at Stage 1 who faces schedules rs(p, i) chooses an industry in which the expected utility

is maximized, and that expected utility hinges only on rs(p, i).

Proof of Proposition 4

Begin at period 2. There are four distinct cases to consider. (i) If wa ≤ Sι < Sι or

Sι < wa < Sι, any equilibrium has the feature that the worker leaves and is paid Sι by

the new employer. (ii) If Sι < Sι < wa, it is still desirable that the worker leaves, but

now that requires a severance pay of at least wa − Sι to make up for the lost wage. Any

equilibrium has the feature that the incumbent offers exactly this severance pay and the

35

worker departs. (iii) If Sι < Sι ≤ wa or Sι ≤ wa ≤ Sι, the worker stays and exerts effort

e = 1. (iv) If wa < Sι < Sι, the worker ought to stay, but has an attractive outside

option. We know from the analysis of Stage 2b that the equilibrium renegotiation offer

is given by equation (15) Note that all the ensuing payments are consistent with ra and

that the second-period assignments are optimal.

It remains to prove that there exists at least one solution to E[ra(w)] = E[Sι] and

that the largest solution to this equation will be the equilibrium outcome. Note that

w = 0 implies positive expected profit to the firm, and competitive bidding at stage 1

raises the wage above this level. Since r(w) is continuous and unbounded above, there

exists at least one solution in which firm profit is 0 and the equation is satisfied. Define

the highest solution wa (there might exist several solutions if an increase in w reduces

the probability of renegotiation at date 2). It only remains to prove that wa is also the

salary that will be offered in equilibrium. For this, observe that wa is the salary that

gives the initial employer zero expected profit and the worker the highest fixed wage (no

other employer obtains a profit given this contract and the worker’s behavior); thus there

is no other competitive outcome.

Proof of Proposition 5

Keep the outcome in period 1 fixed and arrange the remunerations ra and rs in an

increasing order. Denote the resulting probability densities ha(r) and hs(r) respectively,

and let Ha and Hs denote the corresponding cumulative distribution functions. We have

that (i) E[rs] = E[ra], (ii) rs ≤ ra for max{Sι, Sι + βTι

}< wa, and (iii) rs > ra for

all other values of p (indeed, in these states rs(p) = ra(p) + ws). Hence, there exists a

value, w such that Ha(w) = Hs(w), Ha(w) ≤ Hs(w) for w < w and Ha(w) ≥ Hs(w) for

w > w.30 It follows that ∫ w

0

Hs(r)−Ha(r) dr ≥ 0,

where w is the maximum wage (highest feasible p and a). Because wa > ws, the inequality

is strict at least for the lowest possible spot contract remuneration, r = ws. By the

definition of second-order stochastic dominance we have that the distribution of Ha(r)

dominates Hs(r).

Proof of Proposition 7

The proof follows that of Proposition 4, with two exceptions.

30The single-crossing property of the cumulative distribution functions is not necessary for showingsecond-order stochastic dominance, but it is sufficient.

36

First, the contract no longer needs to be renegotiated when the best offer comes from

industry ι; this modification is immediate.

Second, we need to check that it is never profitable to pay above the best intra-

industry outside option, that is, to set r(p) > Sι(p). The potential benefit from doing so

is to prevent renegotiation (and thus constrain rent-sharing) when pιaι > m(p) > rc(p).

This benefit is increasing in the employee’s bargaining power, so it is sufficient to show

that it is unprofitable to raise contracted pay in case β = 1. A contracted pay raise is

unprofitable if realized pay goes up in “good” states (due to worse insurance); we now

show that it does so.

Suppose the wage offer at date 0 for a given pι is increased from x = pιaι(1 + θκ) to

x + ε. In case m > x + ε, the wage will be renegotiated in the same way as before, and

the adjustment does not matter. Otherwise, the realized wage increases in case m < x

and decreases in case m ∈ (x, x + ε). To be precise, with β = 1, the change in expected

wage payments is

G(x)ε+ [G(x+ ε)−G(x)] [pιaι(1 + θκ) + ε− pιaι(1 + κ)]

= G(x)ε+ [G(x+ ε)−G(x)] [x+ ε− z x] (23)

= G(x+ ε) [(1− z)x+ ε]−G(x)(1− z)x. (24)

Differentiation with respect to ε yields

g(x+ ε) [(1− z)x+ ε] +G(x+ ε).

Thus, the expected wage increases if

G(x+ ε)

g(x+ ε)> (z − 1)x− ε,

which is implied by Assumption 3.

Proof of Proposition 8

The proof is similar to the proof of Proposition 5. Keep the outcome in period 1 fixed.

Arrange the remunerations ra and rc in an increasing order. Denote the resulting proba-

bility densities ha(r) and hc(r) respectively, and let Ha and Hc denote the corresponding

cumulative distribution functions. We have that (i) E[rc] = E[ra], (ii) ra ≤ rc for

max{Sι, Sι + βT (p)

}< wc, and (iii) ra ≥ rc for all other values of p. Note that ra > rc

whenever the contract is renegotiated under salary contract and not under simple perfor-

mance contract (i.e., when I = 1 in (21)). This happens with positive probability because

37

(17) fails. Hence, there exists a value, w0 such that Ha(w0) = Hc(w

0), Hc(w) ≤ Ha(w)

for w < w0 and Hc(w) ≥ Ha(w) for w > w0. It follows that∫ w

0

Ha(r)−Hc(r) dr ≥ 0,

where w is the maximum wage (highest feasible p and a). Because wc > wa, the inequality

is strict for at least the lowest possible spot contract remuneration, r = wa. By the

definition of second-order stochastic dominance we have that the distribution of Hc(r)

dominates Ha(r).

38